Learned Search Parameters For Cooperating Vehicles using Gaussian Process Regressions

Unmanned vehicles are capable of working as teams to accomplish a wide variety of mission objectives, such as searching for and tracking targets. In this paper, vehicle search paths are dictated by a joint cost function which maximizes the reward earned from partitioned sections across the search area. In previous work, these rewards were assigned based on the elapsed time since the section had last been searched. This approach is effective in rewarding vehicles to search out areas which haven't been visited in a long time, yet it lacks the ability to weight grid cells differently based on the probability that targets will be in that section. This paper proposes a method of using accumulated knowledge of the average density of targets within an area, along with a Gaussian process regression to assign rewards. Vehicles then choose paths that are more likely to find targets rather than seeking areas which have not been searched recently. Through numerical simulations we show that this method increases the number of targets seen by cooperating UAVs and provides an accurate estimate of target density within a search area.